Tools
ToolsEl-Naggar, Nadine, Ryzhikov, Andrew, Daviaud, Laure, Madhyastha, Pranava and Weyde, Tillman (2023) Formal and Empirical Studies of Counting Behaviour in ReLU RNNs. In: The 16th International Conference on Grammatical Inference. Proceedings of Machine Learning Research, 217 . UNSPECIFIED, pp. 199-222.
Preview |
PDF (accepted manuscript)
- Accepted Version
Download (1MB) | Preview |
Abstract
In recent years, the discussion about systematicity of neural network learning has gained renewed interest, in particular the formal analysis of neural network behaviour. In this paper, we investigate the capability of single-cell ReLU RNN models to demonstrate precise counting behaviour. Formally, we start by characterising the semi-Dyck-1 language and semi-Dyck-1 counter machine that can be implemented by a single Rectified Linear Unit (ReLU) cell. We define three Counter Indicator Conditions (CICs) on the weights of a ReLU cell and show that fulfilling these conditions is equivalent to accepting the semi-Dyck-1 language, i.e. to perform exact counting. Empirically, we study the ability of single-cell ReLU RNNs to learn to count by training and testing them on different datasets of Dyck-1 and semi-Dyck-1 strings. While networks that satisfy the CICs count exactly and thus correctly even on very long strings, the trained networks exhibit a wide range of results and never satisfy the CICs exactly. We investigate the effect of deviating from the CICs and find that configurations that fulfil the CICs are not at a minimum of the loss function in the most common setups. This is consistent with observations in previous research indicating that training ReLU networks for counting tasks often leads to poor results. We finally discuss implications of these results and possible avenues for improving network behaviour.
| Item Type: | Book Section |
|---|---|
| Additional Information: | Funding Information: We thank the anonymous reviewers for their comments and constructive feedback. Laure Daviaud has been supported for this work by the EPSRC grant EP/T018313/1. Andrew Ryzhikov is partially supported by the EPSRC grant EP/T018313/1 and by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant agreement No. 852769, ARiAT). |
| Uncontrolled Keywords: | automata theory,counting,dyck-1,relu,rnn,software,artificial intelligence,control and systems engineering,statistics and probability,2* ,/dk/atira/pure/subjectarea/asjc/1700/1712 |
| Faculty \ School: | Faculty of Science > School of Computing Sciences |
| UEA Research Groups: | Faculty of Science > Research Groups > Algebra, Logic & Number Theory |
| Related URLs: | |
| Depositing User: | LivePure Connector |
| Date Deposited: | 01 Aug 2023 08:30 |
| Last Modified: | 11 Nov 2024 09:30 |
| URI: | https://ueaeprints.uea.ac.uk/id/eprint/92740 |
| DOI: |
Downloads
Downloads per month over past year
Actions (login required)
 |
View Item |